We assert loading that page. The reason is that Windows tells us that fCharPos is false for the first character in a text run, i.e., it's telling us the first character is inside a cluster. We can't handle that because for our purposes clusters must not span text runs. The fix is to always treat the first character as the start of a cluster. Note that with the new textframe, textruns can span text node boundaries so this will be an issue less often, but it can still happen if someone puts a random Arabic character in its own paragraph or in a span with its own font.

checked in.